Fortran 90/95 and Computational Physics

Fortran 90/95 and ComputationalPhysicsJens Hjörleifur Bárðarsonjensba@raunvis.hi.isUniversity of IcelandFortran 90/95 and Computational Physics – p.1/22

OverviewWhat is Fortran?Why Fortran?Some Important ThingsSummaryFortran 90/95 and Computational Physics – p.2/22

What is Fortran 90?Fortran 90/95 and Computational Physics – p.3/22

The OriginA team lead by John Backus developedFortran, FORmula TRANslation System, in1954, one of the earliest high-level languages.Fortran 90/95 and Computational Physics – p.4/22

The OriginA team lead by John Backus developedFortran, FORmula TRANslation System, in1954, one of the earliest high-level languages.1966: The first ever standard for a programming language:Fortran 66New standard 1978: Fortran 77The need to modernise the language → Fortran 90/95Fortran 90/95 and Computational Physics – p.4/22

Fortran 90http://csep1.phy.ornl.gov/pl/pl.htmlFortran 90/95 and Computational Physics – p.5/22

Why Fortran 90?Fortran 90/95 and Computational Physics – p.6/22

How does F90 compare?http://csep1.phy.ornl.gov/pl/pl.htmlOne of the ultimate goals of F90 is that the code must beefficientFortran 90/95 and Computational Physics – p.7/22

Numerical LibrariesFortran has been widely used by scientist and engineers formany years and therfore many algorithms to use innumerical calculations already exist.These have been collected in number of numerical libraries,some open (e.g. SLATEC http://www.netlib.org/slatec/ andNumerical Recipes http://www.nr.com/)and some that cost (e.g. NAG http://www.nag.co.uk).Fortran 90/95 and Computational Physics – p.8/22

Some F90 FeaturesFortran 90/95 and Computational Physics – p.9/22

The ConstructsF90 has many familiar constructs:IF (expr) ...Fortran 90/95 and Computational Physics – p.10/22

The ConstructsF90 has many familiar constructs:IF (expr) ...IF (expr) THEN...END IFFortran 90/95 and Computational Physics – p.10/22

The ConstructsF90 has many familiar constructs:IF (expr) ...IF (expr) THEN...END IFDO i = 1, n...END DOFortran 90/95 and Computational Physics – p.10/22

The ConstructsF90 has many familiar constructs:IF (expr) ...IF (expr) THEN...END IFDO i = 1, n...END DOOther forms of the DO constructFortran 90/95 and Computational Physics – p.10/22

The ConstructsF90 has many familiar constructs:IF (expr) ...IF (expr) THEN...END IFDO i = 1, n...END DOOther forms of the DO constructCASEFortran 90/95 and Computational Physics – p.10/22

Numeric Kind ParameterisationProgram test_kindImplicit noneReal :: a! selected_real_kind([p][,r]) p = precision, r = rangeInteger, parameter :: long = selected_real_kind(9,99)Real(long) :: ba = 1.7; b = 1.7_longPrint *, a,kind(a), precision(a), range(a)Print *, b,kind(b), precision(b), range(b)b = 1.7;b = 1.7D0;print *, bprint *,bEnd Program test_kindFortran 90/95 and Computational Physics – p.11/22

IMPLICIT NONEStrong typing: all typed entities must have their typesspecified explicitlyBy default an entity in Fortran that has not been assigned atype is implicitly typed, e.g. entities that begin with i,j, ... areof type integer → dangerous source of errors(Legend has it that error of this type caused the crash of the American Space Shuttle)The statement IMPLICIT NONE turns on strong typing andits use is strongly recommendedFortran 90/95 and Computational Physics – p.12/22

Modules - Simple ExampleMODULE constantsIMPLICIT NONEINTEGER, PARAMETER :: long = SELECTED_REAL_KIND(15,307)REAL(long), PARAMETER :: pi = 3.14159265358979324D0END MODULE constantsPROGRAM module_exampleUSE constantsIMPLICIT NONEREAL(long) :: aa = 2D0*piprint*, aEND PROGRAM module_exampleFortran 90/95 and Computational Physics – p.13/22

Modules - Another ExampleMODULE circleUSE constantsIMPLICIT NONECONTAINSFUNCTION area(r)REAL(long), INTENT(IN) :: rREAL(long):: areaarea = 2D0*pi*rEND FUNCTION areaFUNCTION circumference(r)REAL(long), INTENT(IN) :: rREAL(long):: circumferencecircumference = pi*r**2END FUNCTION circumferenceEND MODULE circleFortran 90/95 and Computational Physics – p.14/22

Modules - Another Example - cont.PROGRAM module_example2USE constantsUSE circleIMPLICIT NONEREAL(long) :: r, A, Cr = 2A = area(r)C = circumference(r)print*, A, CEND PROGRAM module_example2Fortran 90/95 and Computational Physics – p.15/22

Array FeaturesPROGRAM arrayUSE constantsIMPLICIT NONEREAL(long), DIMENSION(10,10) :: aREAL(long), DIMENSION(5,5) :: b,cREAL(long):: da = 1D0; b = 2D0c = MATMUL(a(1:5,6:10),b)c = c + bd = SUM(c)print*, dEND PROGRAM arrayFortran 90/95 and Computational Physics – p.16/22

External SubroutinesSUBROUTINE area_rectangle(l,b,A)USE constantsIMPLICIT NONEREAL(long), DIMENSION(:,:), INTENT(IN) :: l,bREAL(long), DIMENSION(size(l,1), size(l,2)) :: AA = l*bEND SUBROUTINE area_rectangleFortran 90/95 and Computational Physics – p.17/22

External Subroutines - cont.PROGRAM subr_exampleUSE constantsIMPLICIT NONEINTERFACESUBROUTINE area_rectangle(l,b,A)USE constantsIMPLICIT NONEREAL(long), DIMENSION(:,:), INTENT(IN):: l,bREAL(long), DIMENSION(size(l,1), size(l,2)), INTENT(OUT) :: AEND SUBROUTINE area_rectangleEND INTERFACEREAL(long), DIMENSION(2,2) :: l,b,Al = 1D0; b = 2D0CALL area_rectangle(l,b,A);print*, AEND PROGRAM subr_exampleFortran 90/95 and Computational Physics – p.18/22

External Subroutines - cont.External subroutines are implicitly interfaced whilemodule subroutines are explicitly interfacedFortran 90/95 and Computational Physics – p.19/22

External Subroutines - cont.External subroutines are implicitly interfaced whilemodule subroutines are explicitly interfacedExternal subroutines can be made explicitly interfacedby the use of an interface blockFortran 90/95 and Computational Physics – p.19/22

External Subroutines - cont.External subroutines are implicitly interfaced whilemodule subroutines are explicitly interfacedExternal subroutines can be made explicitly interfacedby the use of an interface blockGrouping related procedures and parameters intomodules is good programmingFortran 90/95 and Computational Physics – p.19/22

External Subroutines - cont.External subroutines are implicitly interfaced whilemodule subroutines are explicitly interfacedExternal subroutines can be made explicitly interfacedby the use of an interface blockGrouping related procedures and parameters intomodules is good programmingWe imagine subprogram libraries being written as sets of externalsubprograms together with modules holding interface blocks forthem. Metcalf & ReidFortran 90/95 and Computational Physics – p.19/22

SummaryFortran 90/95 and Computational Physics – p.20/22

SummaryFortran has from the beginning been designed fornumerical calculationsThe Fortran 90 standard modernised the languageArray features make F90 especially attracting fornumerical workFortran is fastFortran 90/95 and Computational Physics – p.21/22

ResourcesCSEP. Fortran 90 and Computational Science.Technical report, Oak Ridge National Laboratory, 1994http://csep1.phy.ornl.gov/CSEP/PL/PL.htmlThe Liverpool Fortran 90 courses homepagehttp://www.liv.ac.uk/HPC/F90page.htmlMichael Metcalf and John Reid. Fortran 90/95 explained,second edition. Oxford, 1999Chivers and Sleightholme. Introducing Fortran 95. Springer,2000Brainerd, Goldberg and Adams. Programmer’s Guide toFortran 90, third edition. Springer, 1996dbforums.lang.fortran http://dbforums.com/f132/Fortran 90/95 and Computational Physics – p.22/22